|
|||||||||||||||||||||||||||||||||||
 |
 |
||||||||||||||||||||||||||||||||||
|
|
|||||||||||||||||||||||||||||||||||
|
The International Solid Earth Research Virtual Observatory Geoffrey Fox and Marlon Pierce - Community Grids Lab Indiana University We describe the architecture and initial implementation of the International Solid Earth Research Virtual Observatory (iSERVO). This has been prototyped within the USA as SERVOGrid and expansion is planned to Australia, China, Japan and other countries. We base our design on a globally scalable distributed “cyber-infrastructure” or Grid built around a Web Services-based approach consistent with the extended Web Service Interoperability approach. The Solid Earth Science Working Group of NASA has identified several challenges for Earth Science research. In order to investigate these, we need to couple numerical simulation codes and data mining tools to observational data sets. This observational data are now available on-line in internet-accessible forms, and the quantity of this data is expected to grow explosively over the next decade. We architect iSERVO as a loosely federated Grid of Grids with each country involved supporting a national Solid Earth Research Grid. The national Grid Operations, possibly with dedicated control centers, are linked together to support iSERVO where an International Grid control center may eventually be necessary. We address the difficult multi-administrative domain security and ownership issues by exposing capabilities as services for which the risk of abuse is minimized. We support large scale simulations within a single domain using service-hosted tools (mesh generation, data repository and sensor access, GIS, visualization). Simulations typically involve sequential or parallel machines in a single domain supported by cross-continent services. We use Web Services implement Service Oriented Architecture (SOA) using WSDL for service description and SOAP for message formats. These are augmented by UDDI, WS-Security, WS-Notification/Eventing and WS-ReliableMessaging in the WS-I+ approach. Support for the latter two capabilities will be available over the next 6 months from the NaradaBrokering messaging system. We augment these specifications with the powerful portlet architecture using WSRP and JSR168 supported by such portal containers as uPortal, WebSphere, and Apache JetSpeed2. The latter portal aggregates component user interfaces for each iSERVO service allowing flexible customization of the user interface. We exploit the portlets produced by the NSF NMI (Middleware initiative) OGCE activity. iSERVO also uses specifications from the Open Geographical Information Systems (GIS) Consortium (OGC) that defines a number of standards for modeling earth surface feature data and services for interacting with this data. The data models are expressed in the XML-based Geography Markup Language (GML), and the OGC service framework are being adapted to use the Web Service model. The SERVO prototype includes a GIS Grid that currently includes the core WMS and WFS (Map and Feature) services. Users interact with remote services through a Web browser portal that is run by the User Interface Server (UIS) and maintains client proxies. These proxies are responsible for generating the SOAP messages appropriate to the particular services’ WSDL descriptions and for receiving the responses from the services. The UIS and most services are implemented in Java using the Apache Axis toolkit, with C++ services using gSOAP for simple remote visualization. A typical interaction involves the user selecting a code through the portal, setting up an input file in part through interactions with databases (such as the QuakeTables Fault Database), invoking the code and monitoring its progress, and having the output visualized through various third party tools of varying sophistication. These interactions are based on a dataflow model: services communicate by exchanging data files, which must be pulled from one server to another. Basic iSERVO services include remote command execution, file upload and download, and host-to-host file transfer. We do not directly alter the geophysical applications included in the portal but instead follow a “proxy wrapping” approach. To support distributed service orchestration, we use a simple “workflow” service based on the Apache Ant project. iSERVO applications work with many different data sources, and we have developed GML-based services to automate the filtering of data between application services. We have designed a set of high performance streaming protocols that are SOAP compatible. These can exploit fast transports such as parallel TCP and UDP supported by NaradaBrokering. The basic SOAP header and transport protocol is negotiated for any message stream linking services. This phase uses slow conventional SOAP messaging over HTTP. Subsequent messages in the stream only transport changes in the SOAP header including those needed for WS-ReliableMessaging and a token referencing the negotiation and its metadata. This approach will support both high volume sensors and linking visualization and simulation services. We will follow the best practice in the Grid and Web Service field and will adapt our technology as appropriate. For example, we expect to support services built on WS-RF when is finalized and to make use of the database interfaces OGSA-DAI and its WS-I+ versions.Finally, we review advances in Web Service scripting (such as HPSearch) and workflow systems (such as GCF) and their applications to iSERVO. |
|||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||
